https://www.thenakedscientists.com/articles/science-news/machine-learning-turns-thoughts-words

Reading the pattern of activated brain cells presents great opportunities. Already brain scientists can generally correlate areas of the brain to types of activities, however, machine learning presents the opportunity to learn specific brain cell activation patterns that relate to specific activities.

Neural nets are the way that machine learning figures things out by simulating a network of neurons in a human brain.
With today’s technology, a major limitation is the amount of “neurons” that exist in a standard sized machine learning system.  According to Moore’s law,  computer processing power continually increases. Machine learning systems will have the capability of matching or surpassing the number of neurons within the human brain in the next few years.
We will then be able to simulate human thinking and decision-making processes using an artificial brain.  Many psychological concepts such as “unconscious bias” and “confirmation bias” will be modelled using artificial brains.  This will increase our understanding of human thought processes. Exciting times for psychological studies.

As my knowledge about machine learning increases, I’m understanding more and more that machine learning will be a slow but steady evolution, rather than revolution.
Before training data is submitted to the machine learning algorithm to learn, the data must be analysed and “cleaned” by a human first.  There is not yet a machine learning system that can eliminate some of the dirty data that needs to be cleaned from a dataset before the machine learning system is trained using the data.
The types of jobs that will be replaced by machine learning algorithms will be jobs that require simple classification tasks – jobs that require a human to look at similar things to make a judgement as to what category the thing belongs to.  These jobs will be replaced first.
While the machine learning software that is available now is a big step forward, machines cannot yet “learn how to learn”, or deal with outliers in a data set.

How do babies learn? Trial and error.  There was once a theory that the ability to walk, talk and were unlocked over time during a baby’s development.  It is now known that babies try micro muscle movements and get feedback on what is working.  They then string the micro movements together into patterns so that their mobility gets better and better.
How will machines learn in the future? Trial and error.  The feedback loop.  Machine learning does not give exact results, but rather allows the machine to provide the best guess, based on training data and feedback, as to what the result is.  As demonstrated in this video, machines will learn how to walk using trial and error.

Here we go.  The first I heard of machine learning advanced image recognition was when I heard that machine learning algorithms can detect melanomas better than a panel of experts.
As is often the case, technology developed for high-brow uses spreads to all kinds of uses.  This now includes pizza.  Domino’s Pizza will use machine learning to look through a video camera at each pizza made.  The machine learning algorithm can compare the pizza being made to what the algorithm has been trained to identify as the perfect pizza.  The machine learning algorithm will no doubt be able to give each pizza made a score and notify the manager whether the pizza passes minimum standards.
As with other fast food outlets, consistency is key.  Machine learning will not be able to taste the pizza, but it will be able to identify where the pizza doesn’t look right.  A win for pizza lovers everywhere!
 
 

Now that the Australian government’s My Health Record is opt out, it will most probably cover the majority of Australians.  There have been many concerns raised. In order to opt-out, a person must go through a multi-step process to identify themselves with a drivers’ licence and Medicare card.  Most people will not be bothered to go through this process.
A critical privacy issue with My Health Record is that each person cannot see who has accessed their health record.  The lack of this visibility is a critical issue when the privacy of data is concerned.
In the future, health trends, just like DNA services such as 23andme from anonymised de-identified data will be able to be learnt from this data. Machine learning will make the analysis faster.